Most research in efficient speech coding concentrated for many years on algorithms which produced a fixed bit rate. Fixed bit rates are, however, not a requirement for modern packet-based telecommunications and computer networks as well as voice storage applications. This makes the implementation of speech coders feasible, which adapt to the changing properties of speech for improved efficiency or quality. Variable bit-rate coding has therefore become the focus of considerable research activity in recent years [A. Gersho and E. Paksoy, ‘‘Variable Bit Rate Coding,’’ Signal Processing VII, Theories and Applications, EUSIPCO 1994, pp. 1169–1173; L. Zhang et al., ‘‘A CELP Variable Rate Speech Codec with Low Average Rate,’’ ICASSP 1997, pp. 735–738; B. C. Xydeas, ‘‘Source Driven Variable Bit Rate Prototype Interpolation Coding,’’ ICASSP 1996, p. 220; B. Shen et al., ‘‘A Robust Variable-Rate Speech Coder,’’ ICASSP 1995, p. 249]. While all of these coders focus on classification procedures for efficient bit allocation, the analysis frame size remains mainly static. The coder featured in this paper adapts the analysis frame size and bit allocation according to the pitch of the signal. A pitch frame extractor and classifier at the front end feeds the detected frame into the speech coding back end, which uses traditional CELP-based techniques [B. S. Schroeder et al., ‘‘CodeExcited Linear Prediction (CELP): High Quality Speech at Very Low Bit Rates,’’ ICASSP 1985, pp. 937–940]. This approach allows the reduction of the bit rate from a constant 4800 bps (bits per second) with original CELP to typically 3200–4000 bps, depending on the speech situation. This figure is even reduced to typically 2600–3200 with the inclusion of Voice Activity Detection. The speech quality equals the quality of the original CELP coder.